From 394fa3c11873ae5355b8c34e4c4a54430f3935c3 Mon Sep 17 00:00:00 2001
From: Michael SCHAL <michaelschal@gmail.com>
Date: Sat, 29 Nov 2025 02:00:42 +0100
Subject: [PATCH] feat: implement AI service for data extraction with dynamic
 configuration and robust response handling.

---
 app/services/ai_service.py | 20 ++++++++++-
 verify_ai_fallback.py      | 70 ++++++++++++++++++++++++++++++++++++++
 2 files changed, 89 insertions(+), 1 deletion(-)
 create mode 100644 verify_ai_fallback.py

diff --git a/app/services/ai_service.py b/app/services/ai_service.py
index 68d62ae..8f1428f 100644
--- a/app/services/ai_service.py
+++ b/app/services/ai_service.py
@@ -270,7 +270,25 @@ class AIService:
             f"(temp={config['temperature']}, max_tokens={config['max_tokens']}, "
             f"timeout={config['timeout']}s, key={sanitized_key})"
         )
-        response = await acompletion(**kwargs)
+        
+        try:
+            response = await acompletion(**kwargs)
+        except Exception as e:
+            # Check for BadRequestError (often due to unsupported parameters like response_format)
+            # We check string representation or type if possible, but litellm errors can be tricky
+            # The user log showed: litellm.exceptions.BadRequestError
+            is_bad_request = "BadRequestError" in str(type(e).__name__) or "400" in str(e)
+            
+            if is_bad_request and "response_format" in kwargs:
+                logger.warning(
+                    f"Model {config['model']} likely does not support JSON mode. "
+                    f"Retrying without response_format. Error: {e}"
+                )
+                del kwargs["response_format"]
+                # Retry without json mode
+                response = await acompletion(**kwargs)
+            else:
+                raise e
 
         content = response.choices[0].message.content
         return content or ""
diff --git a/verify_ai_fallback.py b/verify_ai_fallback.py
new file mode 100644
index 0000000..c662953
--- /dev/null
+++ b/verify_ai_fallback.py
@@ -0,0 +1,70 @@
+import asyncio
+import logging
+from unittest.mock import MagicMock, patch
+from app.services.ai_service import AIService
+
+# Configure logging
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+# Mock BadRequestError since we might not have litellm installed in the environment running this script
+class MockBadRequestError(Exception):
+    pass
+
+async def verify_fallback():
+    logger.info("Starting AI fallback verification...")
+    
+    # Mock config
+    config = {
+        "provider": "openrouter",
+        "model": "google/gemini-2.5-flash-image-preview",
+        "api_key": "fake-key",
+        "api_base": "https://openrouter.ai/api/v1",
+        "temperature": 0.1,
+        "max_tokens": 100,
+        "timeout": 30,
+    }
+    
+    # Mock success response
+    mock_response = MagicMock()
+    mock_response.choices = [MagicMock(message=MagicMock(content='{"price": 10.0}'))]
+    
+    # Patch acompletion
+    with patch("app.services.ai_service.acompletion") as mock_acompletion:
+        # Setup side effect: First call raises BadRequestError, second call succeeds
+        mock_acompletion.side_effect = [
+            MockBadRequestError("400 Bad Request: The request is not supported by this model."),
+            mock_response
+        ]
+        
+        try:
+            # We need to patch the exception check in the code if we can't import the real exception
+            # But the code uses string check "BadRequestError" in str(type(e).__name__)
+            # So MockBadRequestError should work if we name it right or if the code checks "400"
+            
+            # Actually, let's just run it and see if our logic catches it.
+            # The code checks: is_bad_request = "BadRequestError" in str(type(e).__name__) or "400" in str(e)
+            # Our MockBadRequestError has "400" in the message, so it should be caught.
+            
+            logger.info("Calling call_llm...")
+            result = await AIService.call_llm("test prompt", "data:image/...", config)
+            
+            logger.info(f"Result: {result}")
+            
+            # Verify calls
+            assert mock_acompletion.call_count == 2
+            logger.info("SUCCESS: acompletion was called twice (retry worked)")
+            
+            # Verify second call didn't have response_format
+            call_args = mock_acompletion.call_args_list[1]
+            kwargs = call_args.kwargs
+            if "response_format" not in kwargs:
+                logger.info("SUCCESS: Second call did not have response_format")
+            else:
+                logger.error("FAILURE: Second call still had response_format")
+                
+        except Exception as e:
+            logger.error(f"FAILURE: Exception raised: {e}")
+
+if __name__ == "__main__":
+    asyncio.run(verify_fallback())